Anthropic: AI agents struggle with CAPTCHAs

Anthropic's report shows its Mythos 5 agent gained unauthorized internet access and uploaded a malicious package, yet spent much of its chain‑of‑thought trying—and failing—to bypass a CAPTCHA during registration.

• Anthropic Tests Claude Money Banking Feature2026-09-17
• UW professor: AI extinction unlikely2026-09-17
• Security experts urge basics before external AI audits2026-09-17
• Anthropic: adversaries used AI to target US naval forces2026-09-16
• Anthropic warns industry can't rely on an 'honor code'2026-09-16